Computer Vision – Masky
Masky is an AI face detection model which we’ve created in order to decrease chances of spreading COVID in public spaces. We were motivated to protect our community and especially those who are more vulnerable to coronavirus. Masky has been trialed at multiple stores, and people were positively surprised by the friendly reminder of our favorite lion.
The Results:
- 800 images used
- 5+ stores tested
- 500+ alerted
Masky was designed as a friendly lion, which encourages people entering public spaces, such as supermarkets or shops, to wear masks and do so correctly. Our AI lion started its journey as a simple set of data, which we took from Kaggle. We took a set of 800 pictures with different combinations- with masks, without masks or masks worn incorrectly.
TechTalk
To be more specific and technical, the model we used is called Faster RCNN, which is capable of detecting and differentiating any objects. For example, in the past, our clients used a similar model to detect builders who are not wearing helmets on construction sites. In the case of Masky Faster RCNN was applied to detect people’s faces, as they were entering public spaces.
We then had to swap the last layer of the model (predictive) and retrain it by applying transfer learning. It took us two hours to train the model through the application of cloud GPU. Our little yet important mission was accomplished- the retrained model was able to detect both masks and faces.
Putting it all together
Our final step was to gather all the necessary equipment, build Masky and ensure that it functions effectively. Necessary equipment includes a TV, TV stand, webcam and processing unit. The webcam was placed on the TV, so it could effectively detect and take pictures of people. The pictures were then uploaded to our API server, processed by object detection model and returned to the website.
Reactions
People respond in multiple ways to a reminder like ours, so we had to ensure that people were not put off by it. Taking this in consideration we decided that a flat design style combined with a friendly lion exhibiting various emotions would be the best way to convey the reaction. So far we have developed 3 reactions, for the 3 scenarios that could unfold.
- person is wearing the mask – happy reaction;
- person is wearing a mask incorrectly – silly reaction;
- person is not wearing the mask at all – sad reaction.
The prototype of Masky was developed as a way to encourage people to wear masks correctly, but some of our clients requested an advanced solution. Therefore, we’ve connected Masky with an automatic door, which would only open if Masky indicated that the person entering the public space was wearing a mask correctly.
